Attempt to let nss's SSSE3 files build with clang-cl.

Attempt to let nss's SSSE3 files build with clang-cl.

cl.exe can use intrinsics (like e.g. _mm_shuffle_epi8()) even if the arch
targeted by the compiler doesn't support the SSE level needed by the intrinsic.

clang (and gcc) can't do this, because it converts intrinsics into general
LLVM operations, and the LLVM bitcode is then translated back into assembly
later on, based on the target arch.

So move the file requiring an SSSE3 into a new target that's built with
/arch:AVX. The caller of the function in the new target already checks that
the processor supports AVX before doing the call. This also allows cl.exe to
emit AVX code for the regular C code in that function.

(It's unfortunate that clang-cl deviates fairly heavily from cl in this regard.
But this is very difficult to change, and it's the only larger deviation so
far.)

BUG=82385
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=291603

[wtc, 2014-08-21 23:27:49]

Patch set 1 LGTM. I suggest some changes.

https://codereview.chromium.org/493883002/diff/1/nss.gyp
File nss.gyp (right):

https://codereview.chromium.org/493883002/diff/1/nss.gyp#newcode461
nss.gyp:461: # should be on 'nss'.

Nit: we probably don't need to copy this comment.

https://codereview.chromium.org/493883002/diff/1/nss.gyp#newcode473
nss.gyp:473: 'defines': [

BUG?: I believe we also need these, at least to be safe:

  'MP_API_COMPATIBLE',
  'NSS_DISABLE_DBM',
  'NSS_STATIC',
  'NSS_USE_STATIC_LIBS',
  'RIJNDAEL_INCLUDE_TABLES',
  'SHLIB_VERSION=\"3\"',
  'SOFTOKEN_SHLIB_VERSION=\"3\"',
  'USE_UTIL_DIRECTLY',
  'SHLIB_SUFFIX=\"dll\"',
  'SHLIB_PREFIX=\"\"',
  'SOFTOKEN_LIB_NAME=\"softokn3.dll\"',
  'XP_PC',
  'WIN32',
  'WIN95',
  '_WINDOWS',

https://codereview.chromium.org/493883002/diff/1/nss.gyp#newcode482
nss.gyp:482: 'USE_HW_AES',

Nit: it's safer to add 'INTEL_GCM'. It's actually not necessary because only
rijndael.c tests the INTEL_GCM macro. But you need to inspect the source code to
know this.

https://codereview.chromium.org/493883002/diff/1/nss.gyp#newcode488
nss.gyp:488: 'nss/lib/freebl/ecl',

Do we really need this directory?

[Nico, 2014-08-25 17:35:09]

Thanks, all done. Committing…

https://codereview.chromium.org/493883002/diff/1/nss.gyp
File nss.gyp (right):

https://codereview.chromium.org/493883002/diff/1/nss.gyp#newcode461
nss.gyp:461: # should be on 'nss'.
On 2014/08/21 23:27:49, wtc wrote:
> 
> Nit: we probably don't need to copy this comment.

Done.

https://codereview.chromium.org/493883002/diff/1/nss.gyp#newcode473
nss.gyp:473: 'defines': [
On 2014/08/21 23:27:49, wtc wrote:
> 
> BUG?: I believe we also need these, at least to be safe:
> 
>   'MP_API_COMPATIBLE',
>   'NSS_DISABLE_DBM',
>   'NSS_STATIC',
>   'NSS_USE_STATIC_LIBS',
>   'RIJNDAEL_INCLUDE_TABLES',
>   'SHLIB_VERSION=\"3\"',
>   'SOFTOKEN_SHLIB_VERSION=\"3\"',
>   'USE_UTIL_DIRECTLY',
>   'SHLIB_SUFFIX=\"dll\"',
>   'SHLIB_PREFIX=\"\"',
>   'SOFTOKEN_LIB_NAME=\"softokn3.dll\"',
>   'XP_PC',
>   'WIN32',
>   'WIN95',
>   '_WINDOWS',

Done.

https://codereview.chromium.org/493883002/diff/1/nss.gyp#newcode482
nss.gyp:482: 'USE_HW_AES',
On 2014/08/21 23:27:49, wtc wrote:
> 
> Nit: it's safer to add 'INTEL_GCM'. It's actually not necessary because only
> rijndael.c tests the INTEL_GCM macro. But you need to inspect the source code
to
> know this.

Done.

https://codereview.chromium.org/493883002/diff/1/nss.gyp#newcode488
nss.gyp:488: 'nss/lib/freebl/ecl',
On 2014/08/21 23:27:49, wtc wrote:
> 
> Do we really need this directory?

Yes, I only added directories that were necessary to get the target to compile.

[Nico, 2014-08-25 17:36:37]

The CQ bit was checked by thakis@chromium.org

[commit-bot: I haz the power, 2014-08-25 17:37:41]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/thakis@chromium.org/493883002/20001

[commit-bot: I haz the power, 2014-08-25 17:37:57]

Committed patchset #2 (20001) as 291603

[wtc, 2014-08-25 19:23:04]

Patch set 2 LGTM.